Do your friends know these languages? I'm afraid most people have to use translation tools to understand them all!
People live in different places and use different languages. It is the emergence of translation that makes cross-regional and cross-language communication possible. In order to facilitate communication, translation tools are constantly upgraded, and translation functions are becoming more and more powerful, from word-for-word translation to website, picture and even document translation.
Therefore, in order to provide smarter and more intimate translation services so that everyone can communicate with the wider world, Huawei EMUI has successively launched voice translation, AR translation and full-screen translation. This time, Huawei's long text translation black technology-rolling translation also appeared!
Full-screen translation is not enough, and rolling translation is one in a million.
Do you often encounter such a situation: when writing a thesis, there are a lot of foreign references? Is it dumbfounded to travel abroad? Want to catch up with a foreign hot spot, only to find that you can't read the news at all?
If you want to translate multi-screen content, you need to do full-screen translation repeatedly. With the increase of content length, the complexity of operation will also reduce the coherence of reading. Faced with such a scene, full-screen translation has become "inadequate". At this time, rolling translation is needed to help!
Scroll translation supports the translation of multiple applications, ten languages and multi-screen content. Use the small I voice password to "help me translate the screen" or press and hold the screen with two fingers, and click "full screen translation >; Scroll translation ",you can start your" scroll translation "journey!
Next, let's see how rolling translation completes the translation of long articles!
Wonderful assembly line: text → picture → translation
When you browse a long article and need to translate it, the system will first take a screenshot to form a long screenshot of the original text, then cut, detect, translate, sort, copy and splice the long screenshot, and finally present it to you again with the same picture and translated text. This is called rolling translation.
In this series of steps, the most important thing is how to deal with the long screenshots preliminarily. Then what technology is used to deal with it? Here I have to mention the key technology-OCR technology.
The question is, where is OCR?
OCR, that is, optical character recognition, refers to the process of checking, detecting and recognizing characters, and then converting their shapes into computer characters. The application here is to identify and extract the characters on the long screenshot formed by the original long article, and provide them to the subsequent machine translation, that is, to complete the extraction process of long source characters.
Technically, it is a two-stage algorithm framework for detection and recognition, which involves the recognition ability of Chinese, English, Japanese, Korean, Russian, Spanish, French, German, Italian and Portuguese, and includes several depth models running in NPU (Neural Network Processing Unit).
After the OCR algorithm is called, firstly, the image is enhanced, and after the image adapts to the ideal state, the depth model and post-processing operation are run synchronously by multithreading, and finally the output text recognition result is corrected by NLU (Natural Language Understanding).
The accuracy of translation depends on the accuracy of OCR recognition. The speed of translation depends on the speed at which OCR processes characters. Machine translation of OCR-processed text, reordering and splicing of pictures, and finally presenting the translated long text to users.
This innovative scheme can transform the words in a long article into pictures and then into translation, and has applied for patent protection.
Every word and every picture should be complete.
In rolling translation, OCR text line positioning is innovatively used to realize intelligent image segmentation and splicing, which cleverly avoids the situation that the text or image is truncated when long screenshots are segmented.
You may have such a question, can't the long screenshot formed by the original text be directly translated? Why do we have to subdivide?
This is because the current picture translation only supports the translation of single-screen content, long screenshots will be automatically scaled to the same height as the screen for translation, pictures will be compressed, and the text size will naturally be compressed, which greatly affects the accuracy of translation results.
In order to effectively improve the accuracy of translation, before translation, cut the long screenshot into several short pictures. However, if you simply divide a long screenshot according to the screen height, it is likely to truncate a line of text or an illustration.
In rolling translation, OCR technology can detect the position coordinates of a line of text or a picture. If it is found that the segmentation position is in the middle of a line or a picture, it will move up to the blank position between lines for segmentation.
The width of the picture processed according to this rule is the same as the screen, and the height is equal to or slightly lower than the screen height. Each line of text and each picture are complete. This scheme is ingenious and practical, which ensures everyone's translation experience at the same time.
At present, Huawei mobile phone EMUI 10 version can be experienced! The system automatically scrolls to form long screenshots for translation, which is convenient and fast, and truly translates what you see and scrolls the full text.